Click here to


Are you sure ?

Yes, do it No, cancel

Extended Shell Feature: Influence of Tumor Extension in Distant Metastasis Prediction for Non-Small Cell Lung Cancer

H Hao1*, Z Zhou2 , S Li3 , M Folkert2 , L Yang4 , P Iyengar2 , K Westover2 , J Wang2 , (1) Xidian University, School of Computer Science and Technology, Xi'an, China, (2) UT Southwestern Medical Center, Radiation Oncology, Dallas, Texas, (3) Southern Medical University, Guangzhou, China, (4) China Academy of Chinese Medical Sciences, Beijing, China


(Thursday, 8/2/2018) 7:30 AM - 9:30 AM

Room: Davidson Ballroom B

Purpose: Using an extended shell feature, we aim to: 1) analyze the association between tumor periphery and distant metastasis (DM); and 2) to evaluate the influence of the tumor extension distance beyond gross tumor on the prediction performance of DM for inoperative early stage non-small cell lung cancer (NSCLC) patients.

Methods: The extended shell was constructed from pre-treatment positron emission tomography (PET) images of 48 early stage (IA and IB) NSCLC patients received stereotactic body radiation therapy. The proposed feature, extracted from the periphery voxels beyond tumor edge, is a descriptor of tumor extension where different extended sizes corresponding to different margins added to the gross tumor. The hypothesis behind this feature is that the tumor microenvironment may regulate cancer metastasis and thus the corresponding area — the outer part of tumor edge, can be used for distant failure prediction and its performance may vary with different distances beyond the gross tumor boundary. Support vector machine (SVM) was employed to evaluate the efficacy of the proposed feature with ten different margins added to the tumor boundary ranging from 1 pixel (4 mm) to 10 pixels (40 mm) with 1 pixel increment. The shell feature (without extension) previously proposed by our group was compared with the extended shell features. The performance was assessed by the area under the characteristic curve (AUC), sensitivity, specificity and accuracy.

Results: The predictive power of the extended shell features first increases and then decreases as the extended size increases. The best performance is achieved by the extended shell with 4-pixel (16-mm) margin, with an AUC of 0.89 compared to that of 0.82 with no extension.

Conclusion: Tumor periphery is associated with cancer metastasis. An extension with no more than 16-mm appears to be sufficient to predict DM for early stage NSCLC patients.


Quantitative Imaging, PET, Image-guided Therapy


IM- PET : Quantitative imaging/analysis

Contact Email